Use of Multiple RELAY bridges

[psylsph]

Does anybody have experience using multiple RELAY bridges to join more than 2 clusters, ideally I want to be able to join N remote clusters into a single super cluster using UCP multicast within the local clusters and TCP between them. We having problems trying to get the various remote clusters to see each other. If anyones has done this do you have an example with 3 clusters I could scale up.

Thanks

Stuart

[dsb]

I far as I know it's not supported yet. Here's a thread I started on the JGroups mailing list, take a look at the last response from Bela Ban:

http://old.nabble.com/RELAY-usage-td32062001.html

"Speaking of which: we intend to try out my JBW demo on Infinispan 5.1 / JGroups 3.0 soon, to see if the changes I proposed in 4.2 have been forward ported to 5.1 correctly. The intention is to package this as a feature of the EDG product, so expect a lot more work in this area, better documentation and so on. I also have 2 JIRA issues open for JGroups 3.1, which will make RELAY better, plus I intend to add code to use RELAY between more than just 2 sites."

-David

[psylsph]

Thanks

[belaban]

Yes, to use RELAY between more than 2 sites is not supported atm. However, I intend to look into this in the 3.1 release of JGroups. Incidentally, the topic came up today.

[psylsph]

Do you have a time frame for the release of 3.1? I ask because I'm doing a product evaluation and selection at the moment. I've down selected between Infinispan over JGroups and another product. Infinispan over JGroups looks good, however ideally we need a product that can handle unreliable communication between multiple sites.

Ideally I'd like to use RELAY, although there is always the possibity of running TCP_NIO for the geographically co-locationed sites and then using a RELAY to the further aways sites I suppose.

Stuart

[belaban]

No, I don't have a hard date for the 3.1 release yet. I'd like to release it before next summer, but if there's an important chunk of functionality that's done, I might de-scope 3.1, pushing everything else into 3.2, and release earlier.

I just talked to a guy who needs to use RELAY for a content distribution network a la Akamai (or the MBone, if anyone remembers that). This requires a tree form, where sites can have child sites and so on, hierarchically structured.

If you have requirements, I'd like to hear them, so I can take them into account when tackling multi-site distribution.

Note that I created another issue regarding RELAY today: https://issues.jboss.org/browse/JGRP-1401.

[psylsph]

In terms of approach I have no feeling about whether a round robin of all the sites or a star approach is better. Except that it would be nice to have some kind of semi-automated fall over mechanism. Where for instance if the center node dies the relay co-ordinators hold a list that nominates who reforms the center of the star.